AniBridge Secures Seed Funding to Lower AI Inference Costs

AniBridge, a startup founded by professors from the Korea Advanced Institute of Science and Technology, has received seed investment to develop software that reduces the cost of running artificial intelligence services. The company aims to lower inference expenses by utilizing a combination of different semiconductor types rather than relying solely on graphics processing units. Platum reported that the venture capital firm The Ventures provided the seed funding, with the specific amount remaining undisclosed. The company was also selected for the Tips program by the Ministry of SMEs and Startups based on a recommendation from The Ventures.

The software creates an execution environment that bundles various AI semiconductors, including graphics processing units, neural processing units, and processing-in-memory chips. It automatically assigns inference tasks to the most suitable hardware based on response speed, cost, and model size. The goal is to reduce inference costs by forty to seventy percent compared to using only graphics processing units. The system is compatible with application programming interfaces from OpenAI, Google Gemini, and Anthropic Claude, allowing companies to adopt it without changing their existing development methods. AniBridge is also developing an enterprise AI agent platform.

The founders, Park Jong-se and Kwon Young-jin, have published over one hundred papers in top computer systems conferences and have conducted joint research with Google, Meta, Nvidia, Samsung Electronics, and SK Hynix. Kunle Olukotun, a professor at Stanford University and co-founder of Cerebras Systems, serves as an advisor. The technology is based on the open-source simulator LLMServingSim, which won the best paper award at the International Symposium on Workload Characterization in 2024.

AniBridge has begun collaborating with domestic AI semiconductor companies. In September, it signed agreements with FuriosaAI and HyperXell to co-develop an NPU-based enterprise AI agent platform. The company also received the grand prize in the Kakao AI cultivation project among more than sixty participating firms. The investment and Tips funds will be used for product development and customer validation. The company plans to conduct pilot deployments with two or three customers this year and verify cloud performance with Kakao Cloud before launching the official product in 2027.